Skip to content
Petr Baudis edited this page Sep 11, 2015 · 1 revision

Answer Ranking

Eventually, we need to rank all the candidate questions so that we can output the top.

TODO list issue

TODO why don't I think it's that helpful now (training data size impact)

TODO dataset (AP recall)

Baseline

TODO

Other Ideas

More sophisticated ranking than pointwise, of course.

Pairwise random forest ranker - Bast, Haussmann: More accurate question answering on freebase (2015).

TODO CMU ranking work on OpenEphyra

TODO IBM Watson results